Driving tumor-associated macrophages to a CXCL9^Hi/SPP1^Low phenotype eliminates pancreatic cancer

Summary

Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers, with limited therapeutic options. Recent studies reveal the complex PDAC microenvironment, which fosters interactions between various cell types that contribute to immunosuppression. Among these, tumor-associated myeloid cells with a CXCL9-low, SPP1-high phenotype are particularly abundant and play a significant role in promoting disease progression. We hypothesize that targeting these cells can reprogram the tumor microenvironment and improve survival outcomes. To test this, we develop a myeloid cell-targeting nanoformulation, CANDI470, designed to increase CXCL9 expression and reduce SPP1 levels in tumor-associated myeloid cells. Our results demonstrate that this approach not only enhances immune responses but also achieves remarkable therapeutic efficacy, including cures in murine PDAC models, even when used as a monotherapy. This myeloid cell modulation strategy represents a promising therapeutic avenue for PDAC, offering hope for improved treatment for this otherwise difficult-to-treat malignancy.

Graphical abstract

Highlights

• •The immunosuppressive CXCL9^Lo/SPP1^Hi macrophage phenotype predominates in PDAC
• •Cyclodextrin nanoconstructs carry payloads modulating both CXCL9 and SPP1
• •Myeloid-targeting cyclodextrin nanoformula effectively treats PDAC in murine models

Immunosuppressive CXCL9^Lo/SPP1^Hi myeloid cells are highly abundant in PDAC, adding to the poor prognosis. Using the myeloid cell-targeting cyclodextrin nanoconstruct, Nguyen et al. found that simultaneously modulating CXCL9 and SPP1 through this single, systemically administered nanoformulation is highly effective in treating murine PDAC.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) currently represents the third leading cause of cancer death in the United States¹ and, unlike many other cancers, has a rising incidence.² Most patients present with advanced, unresectable disease, and only 12% survive beyond 5 years. The reasons for these grim statistics are manifold, including the often ill-defined symptoms, the absence of cost-effective screening tools, the lack of accurate diagnostic biomarkers for earlier stage tumors (e.g., the presence of high-grade dysplasia in commonly detected cystic precursor lesions), the difficult-to-access anatomical location, and the fact that only 20% of patients are eligible for curative surgery upon presentation.³ Unfortunately, even in cases where surgical resection can be performed, nearly 3 in 4 patients will develop recurrence within 2 years.⁴

Chemotherapy and radiation have been the primary treatment options for advanced PDAC. In the past decade, two combination regimens have become standard first-line options: FOLFIRINOX, which includes 5-fluorouracil (5-FU), leucovorin, irinotecan, and oxaliplatin, and the combination of gemcitabine with albumin-bound paclitaxel (nab-paclitaxel).^5,6,7 While these chemotherapeutic approaches have significantly increased the survival of patients, long-term outcomes remain poor. Patients who progress on first-line regimens may receive second-line treatment, including liposomal irinotecan combined with 5-FU.⁸ Nevertheless, the 5-year survival rate remains low, underscoring the need for therapeutic approaches that extend beyond chemotherapy.

The tumor microenvironment (TME) of PDAC is distinctive, with stromal cells contributing up to 80% of the entire tumor volume. This dense desmoplastic stroma is primarily composed of cancer-associated fibroblasts (CAFs)⁹ and myeloid-derived immune cells, especially tumor-associated macrophages (TAMs).^10,11 Recent single-cell RNA (scRNA) sequencing studies have further defined the myeloid landscape in human cancers.^{12,13,14,15,16,17} Two insights have emerged from these analyses: (1) SPP1 is highly expressed in macrophage subsets associated with poor outcomes, and (2) SPP1 expression is largely mutually exclusive with markers of more favorable macrophage populations, such as CXCL9. In contrast, TAM expression of M2-associated macrophage markers (e.g., defined by CD163 or MRC1) was not associated with clinical outcome in the same patient cohorts,¹² supporting the notion that M2 polarization reflects an in vitro extreme rather than a physiologically relevant state in vivo.¹⁸ These findings, together with several other studies,^{19,20,21,22,23,24,25} have rekindled interest in SPP1 and CXCL9 as possible therapeutic targets and as opposing prognostic biomarkers.

We hypothesized that the immunosuppressive CXCL9^Lo/SPP1^Hi TAM phenotype in PDAC could be polarized to an immunostimulatory state characterized by CXCL9 expression and other key immunostimulatory mediators. To test this, we used a TAM-targeting cyclodextrin nanoconstruct loaded with a triple small-molecule payload designed to simultaneously modulate CXCL9 and SPP1 through a single systemically administered nanoformulation. The payload consisted of a CSFR1 inhibitor (PLX3397), a TLR7/8 agonist (R848), and a STING agonistic prodrug (MSA-2p). We show that this strategy is highly effective in treating, and in some cases eradicating, PDAC in murine models. Notably, the monotherapeutic formulation, “CANDI470,” achieved cures without chemotherapy and conferred protection against tumor growth upon PDAC re-challenge.

Results

Design of compounded nanoformulation to modulate SPP1 and CXCL9 in TAMs

The PDAC TME is characterized by a high macrophage content, where the vast majority of TAMs are of an immunosuppressive SPP1^Hi phenotype. Furthermore, immunostimulatory programs in these TAMs are often downregulated. Prior research¹² had identified the CSR metric (in which C is CXCL9, S is SPP1, and R is ratio), where a low number (<1) indicates an immunosuppressive phenotype with an adverse outcome and a high number (>1) indicates a desirable immunostimulatory phenotype with a preferred outcome.¹² While prior research had focused on lung, head and neck, and colorectal cancer, here we analyzed available human PDAC datasets (Figure 1). Data from 100 patients show a predominant immunosuppressive phenotype, with over 98% of PDAC patients having a TAM CSR <1. These data are further expanded in Figure S1, showing considerable differences in the human PDAC microenvironment compared to other epithelial cancers.

Figure 1.

Tumor-associated macrophage polarization in PDAC is characterized by a distinctly low CSR compared to other solid tumors

(A) Publicly available single-cell RNA sequencing datasets from 87 PDAC tumors across seven independent cohorts (totaling 96,717 cells) were analyzed. Uniform manifold approximation and projection (UMAP) visualization of all cell types in the data.

(B) Macrophages were subset from the integrated dataset, and their relative expression of CXCL9 and SPP1 was used to compute a CXCL9/SPP1 ratio (CSR) for each cell. UMAP visualization of all macrophages pooled across patients, colored by CSR.

(C) Dot plot showing the mean macrophage CSR for each patient, ordered from lowest to highest.

(D) Violin plots showing the distribution of CSR across 11 tumor types on a log scale. Median values are displayed above each plot. Data were derived from bulk RNA-seq analysis of 5,601 patients across 11 tumor types (The Cancer Genome Atlas [TCGA]). Fib, fibroblasts; Mo, monocytes; Neu, neutrophils; Mac, macrophages; EC, endothelial cells; DC, dendritic cells; pDC, plasmacytoid dendritic cells; PC, plasmocytes; MC, mastocytes; NK, natural killer cells; TuEpi, tumor cells; Esoph, esophagus; HNC, head and neck.

(E) To address CSR in TAMs, we developed a TAM-targeted nanoformulation containing a triple small-molecule payload to modulate CSR polarization. The triple payload consisted of a TLR7/8 agonist (R848), a CSF1R inhibitor (PLX3397), and a STING agonist prodrug (MSA-2p).

Prior research had identified some small molecules and combinations that could reduce either SPP1²⁶ or increase CXCL9²⁷ when delivered specifically to TAMs, but not simultaneously. SPP1 inhibition in TAMs was identified through screens that pointed to a CSF1R pathway, with PLX3397 (pexidartinib) emerging as a lead candidate, particularly in combination with a TLR7/8 agonist (R848).²⁶ Potent CXCL9 inducers were identified separately through screens and shown to include STING agonists and PARP7 inhibitors.²⁷ Here, we developed a three-payload system within a single nanoformulation (CANDI470) containing a CSFR1 inhibitor (PLX3397), a TLR7/8 agonist (R848), and a STING agonistic prodrug (MSA-2p) for systemic delivery to TAMs. The prodrug version of the STING agonist MSA-2 was previously developed for incorporation into cyclodextrins.²⁸ To obtain scalable quantities of the expensive MSA-2 precursor sufficient for in vivo use, we developed an alternative MSA-2p synthetic route (Figure S2). Crucially, the inexpensive starting material, 5,6-dimethoxybenzo{b}thiophene-2-carboxylic acid, was decarboxylated using new conditions (catalytic silver) identified in a brief screening campaign. Thus, MSA-2 could be easily produced in high purity at gram scales.

Physical characterization

The core of CANDI was synthesized by cross-linking succinyl-β-cyclodextrin with lysine to yield a ∼17 nm nanoparticle with a zeta potential of −9.8 mV (Figure 2; Table S1). The cyclodextrin units act as guest-host domains for small-molecule drugs, retaining TAM affinity through their carbohydrate surface. Drug loading studies showed that each mg of CANDI nanoparticle could be loaded with the following amounts of the triple payloads without the need for other excipients or stabilizers: PLX3397, 0.02 mg; R848, 0.035 mg; and MSA-2p, 0.04 mg, for a total of ∼0.1 mg of drug per mg of CANDI (Figure 2). The resultant drug-loaded CANDI470 still had a mean hydrodynamic diameter of 17 nm and a charge of −8.4 mV. Drug release assays showed that >90% of the payload was released within 48 h in vitro. The drug product (CANDI470) remained stable in both frozen (−20°C) and lyophilized forms for long-term storage (Figure 2). Once resolubilized at room temperature, the material remained stable for at least 8 h prior to injection.

Figure 2.

Characterization of CANDI470

(A–C) Synthesis of the cross-linked bis-succinyl-β-cyclodextrin with lysine results in nanoparticles with a mean diameter of ∼17 nm as determined by dynamic light scattering (DLS). (B and C) By transmission electron microscopy (TEM), the particles appear homogenous with a diameter of ∼17 nm (scale bar, 20 nm).

(D) Loading of empty CANDI with triple drugs is possible through cyclodextrin guest-host interactions. Payload drugs are often poorly soluble in water (e.g., the free drug combination in the first image appears turbid) but are readily soluble in CANDI (middle). Unloaded and loaded nanoparticles can be lyophilized for long-term storage (right).

(E) Loading, release, and stability data in aqueous solution. The drug loading for the triple combination (CANDI470) chosen here is ∼0.1 mg total drug per mg nanoparticle, as based on previous screens. The capacity of CANDI for small molecules is ∼0.2 mg/mg CANDI (w/w), depending on payload. The release of loaded small molecules in dialysis experiments is complete by ∼10 h in this non-physiological system. In frozen form, CANDI470 is stable for at least 6 months, the longest time point tested.

Cellular effects

Having identified the appropriate loading for the lead compound (CANDI470), we proceeded to perform experiments in cells (Figure 3; Table S2). CANDI uptake into immortalized murine macrophages (iMACs) peaked at 1 mg/mL, a concentration at which all cells internalized CANDI470. Importantly, at this dose, there was also maximum Cxcl9 induction and Spp1 inhibition as determined by Cxcl9-RFP induction or Spp1-tdTomato suppression in bone marrow-derived macrophages (BMDMs) from reporter mice (Figures 3B and 3C). Empty CANDI showed no appreciable cellular toxicity even at 10-fold higher doses (Figure 3D). At 1–2.5 mg/mL, drug-loaded CANDI showed no major effects on cell viability. For comparison, the free drug cocktail was more cytotoxic with an LD₅₀ of 2.5 mg/mL.

Figure 3.

Cellular effects in macrophages

(A) Cellular uptake of CANDI470-AF647 into iMACs. 100% of all cells internalize the CANDI at concentrations of 1 mg/well (incubation time was 2.5 h).

(B) Dose response for eliciting Cxcl9 in bone marrow-derived macrophages (BMDMs). BMDMs were obtained from Cxcl9-RFP donor mice and incubated with varying amounts of CANDI470 for 24 h, after which RFP fluorescence was measured.

(C–F) Dose response for downregulating Spp1 in BMDMs. BMDMs were obtained from Spp1-tdTomato donor mice and incubated with varying amounts of CANDI470 for 24 h after which tdTomato fluorescence was measured. Cellular toxicity of empty nanoparticle (D), CANDI470 (E), and free drugs (F) following a 2.5 h incubation with cells. The free drug dose refers to the payload-equivalent dose within CANDI470. Note that the empty CANDI has no appreciable toxicity at the concentrations tested. Cellular toxicity at high concentrations is primarily due to small-molecule payloads and is overall lower upon CANDI470 encapsulation. Results are represented as mean ± standard error of the mean (SEM).

To determine the broader pharmacological effects, we next performed RNA sequencing (RNA-seq) of BMDMs exposed to CANDI470 or unloaded empty CANDI (CANDI^E; Figure 4). Interestingly, no effects were observed with CANDI^E at the concentrations tested. With CANDI470, however, many genes were upregulated (Figure 4C) or downregulated (Figure 4D). At the chosen cutoff, we detected 120 upregulated genes, many of which were interferon stimulated. Highly upregulated genes included Cxcl9, Cxcl10, Isg15, Il12, Ccl5, Marco, Cd40, and Il1, but also Nos2 (inducible nitric oxide synthase 2), an enzyme that produces nitric oxide (NO) in response to inflammation and cellular stress. Figure 4C summarizes the top 20 upregulated interferon-stimulated genes (ISGs) with CANDI470 treatment; note that none of them were induced with CANDI^E. Highly downregulated genes included Spp1-associated genes such as Egr1,^29,30 Tnfsf12,³¹ Sort1, Il6ra, Lrp5, and Itga6³² and many M2 genes such as Cd163,^33,34 Trem2,³⁵ Nr4a1,^36,37 Gpx1,³⁸ C1qtnf12,³⁹ Cx3cr1,⁴⁰ Gas6,⁴¹ Lpin1,⁴² Pargγ, Cd93,⁴³ and Mafb.⁴⁴ Spp1 itself did not appear downregulated at the 12-h time point, and subsequent experiments indicate that longer time points (∼2 days) are required to observe these direct Spp1 effects (Figure 6 via intravital imaging in the Spp1 reporter mouse; Figure S3 via qPCR).

Figure 4.

Effect of CANDI470 on BMDMs

(A) BMDMs were obtained from BL6 mice and treated with CANDI470 or CANDI^E as control (1 mg/mL; 2 h). Note the upregulation of immunostimulatory genes, including Cxcl9, Cxcl10, Ccl5, Marco, Cd40, and IL12, among others. Downregulation of Trem2, Cd163, and Slc46a3, among others.

(B) Similar to experiment in (A), but with an empty CANDI. Note that the CANDI particle by itself has no therapeutic efficacy.

(C) List of top 20 upregulated IFNg response genes with CANDI470 treatment ranked by fold change. Control is PBS.

(D) List of 20 downregulated genes with CANDI470 treatment. Many genes are related to M2 and Spp1 polarization. Control is PBS.

Figure 6.

Pharmacodynamics of CANDI470 effects in vivo

Serial intravital imaging (before, immediately after i.v. administration, and at 24 and 48 h) of CANDI470 in KPC-YFP tumor-bearing mice (either expressing Cxcl9-RFP or Spp1-Tdtomato).

(A) Combined colors from (B) and (C) show KPC-YFP tumor cells (green), AF647-labeled CANDI470 (white), and Cxcl9-expressing cells (red).

(B) Fluorescently labeled CANDI was initially observed in the tumor neovasculature but was completely cell associated at 24 and 48 h (see Figure 5 for cell types).

(C) Using Cxcl9-RFP reporter mouse, massive Cxcl9 upregulation was observed in the tumor by 48 h post-injection; most Cxcl9+ cells were also positive for CANDI uptake.

(D) Via Spp1-Tdtomato reporter mouse, Spp1 downregulation was also detected at a later time point. Scale bar, 200 μm. Results are represented as mean ± standard error of the mean (SEM). p values from one-way ANOVA are shown. p < 0.05 was considered significant. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.0001. n.s. denotes not significant.

We next performed qPCR to determine the mRNA levels of Cxcl9 and Spp1 in BMDMs. In these experiments, differentiated BMDMs were treated with CANDI470 at 1 mg/mL. The data show upregulation (an average of 6-fold increase at 12 h; p < 0.001) of Cxcl9 and downregulation of Spp1 (an average of 0.33-fold decrease at 24 h; p < 0.001) (Figure S3A). Time-course experiments in BMDMs showed that therapeutic effects in a pulse-chase experiment peaked within 24 h and returned to baseline at 48–72 h after drug exposure (Figure S4).

Pharmacokinetics

For CANDI470 to accumulate in tumor-associated myeloid cells, it must have a sufficiently long circulation time for systemic delivery. We thus measured its blood half-life in vivo and determined it to exhibit typical bi-functional decay, with a slow half-life of 0.66 h and a fast component of 0.075 h (R²: 0.99; Figure 5). This behavior in murine systems is similar to that of other carbohydrate nanoparticles.^45,46 Twenty-four hours after intravenous (i.v.) administration, CANDI470 showed the highest accumulation in tumors and tumor-draining lymph nodes, with lower amounts in other organs (Figure 5B). While the biodistribution was done with fluorescently labeled CANDI470-AF647 (Table S1), the distribution was also similar to that of ⁶⁴Cu-radiolabeled versions of different cyclodextrin nanoparticle preparations, in which high tumor uptake was observed (e.g., Figure S9 in Fredrich et al.⁴⁷), as well as to the distributions observed in other tumor models.⁴⁸

Figure 5.

Pharmacokinetics and organ and cellular distribution

CANDI470-AF647 was administered via the tail vein at 5 mg/mouse.

(A) Intravital imaging of the mouse ear vein shows a two-phase exponential decay with a slow half-life of 40 min and an initial faster component of 5 min.

(B) Organ distribution of intravenously injected CANDI470 24 h after administration. The highest amounts were observed in PDAC and lymph nodes, similar as in other studies.^47,Results are represented as mean ± standard error of the mean (SEM).⁴⁸

(C) Within the TME of KPC-YFP, CANDI470 was distributed almost exclusively to CD11b-positive TAMs and MHC-II positive antigen-presenting cells (APCs). Each column represents one mouse, and the grayscale reflects the percentage of CANDI470-positive cells in the TME.

(D) Similar data were observed in the Panc02 mouse model of PDAC.

Using flow cytometry (Table S3), we next determined the cell types within tumors in which intravenously injected CANDI470 accumulated (Figure 5). Using two different PDAC models, we observed high accumulation in CD11b+ myeloid cells and, in particular, TAMs and dendritic cells (DCs) (KPC model: 60% of TAMs positive for CANDI470, 74% of DCs positive; PANC2 model: 80% of TAMs positive for CANDI470, 70% of DCs positive). This finding was similar to that observed in other tumor models in which CANDI biodistribution had been studied.⁴⁷ There was negligible uptake in T cells and tumor cells (Figure 5). In summary, these data demonstrate high accumulation of CANDI470 in tumors, particularly in TAMs and DCs, following i.v. administration. We also determined accumulation in non-target organs and cells, finding almost exclusive accumulation in myeloid cells but at much lower levels than in tumors (Figure S5).

To determine whether cellular accumulation in the TME would also translate to macrophage activation, as determined in vitro, we performed single-cell pharmacodynamic studies using intravital microscopy (TAM pharmacodynamics; Figure 6; Figures S6 and S7). These studies were performed in reporter mice, where the temporal expression of Cxcl9 and Spp1 could be determined optically. Imaging demonstrated a time-dependent accumulation in tumor-associated myeloid cells and a significant increase in local Cxcl9 production (from <5 cells to 150 cells per field of view at 48 h; p < 0.001). Similarly, we observed a decrease in Spp1 over time, from ∼75 cells to ∼15 cells per field of view at 85 h (p = 0.0024). These values revert to normal baseline values in about 5 days, hence the dosing schedule of 4 days in Figure 7.

Figure 7.

Therapeutic efficacy

(A and B) Overview of experimental scheme. KPCY tumor-bearing BL6 mice were treated with empty CANDI (n = 9) or CANDI470 (n = 10) on days 0, 4, 8, and 12. Tumor burden was measured via calipers in ectopic tumors or by resection in orthotopic tumors. Dose of CANDI470 (5 mg/mouse).

(B) The control group receiving empty CANDI showed aggressive PDAC growth, similar to that reported for this model.⁴⁹

(C) In contrast, CANDI470-treated animals (4 doses indicated by arrows) showed a flat growth curve, with 7/10 animals showing no residual tumor on day 30 (p < 0.0001, Pearson correlation efficiency analysis).

(D) Survival curve shows therapeutic efficacy of CANDI470 (p < 0.0001, Kaplan-Meier analysis). n = 9–10. Median survival: 22 days for the untreated group versus >3 months for the CANDI470-treated group.

(E) Difference in orthotopic tumor growth on day 12 after treatment. Results are represented as mean ± standard error of the mean (SEM). ∗∗p < 0.005, unpaired t test.

(F) Body weights were measured in both groups and were largely stable during the treatment.

(G) Re-challenge experiment. 1 × 10⁶ KPCY cells were implanted either into naive BL6 mice (black curve; n = 6) or the “cured mice” from (B) (red curve; n = 7). Note the immunologic protection against re-challenge (p = 0.008, unpaired t test).

(H) Similar therapeutic efficacy was observed in an alternative, chemically induced murine PDAC model (Panc02⁵⁰) (p = 0.02, unpaired t test).

Therapeutic efficacy in PDAC

To determine the effects of intravenously administered CANDI470 as a monotherapy, we performed studies in two different PDAC models: KPCY and Panc02 (Figure 7; Table S4). Mice received four i.v. doses of CANDI470 (5 mg; CANDI^E served as a control), beginning 8 days after tumor implantation, and tumor burden was monitored longitudinally by caliper measurements. In addition, we also performed experiments in orthotopically implanted KPCY tumors (Figure 7; Figures S8–S10).

In the KPCY model, control mice treated with CANDI^E showed the expected aggressive tumor growth pattern, consistent with previous reports for this system.⁴⁹ In contrast, CANDI470 markedly suppressed tumor growth, as 9 of 10 treated animals showed no residual tumor on day 30. “Cured” mice were subsequently re-challenged 3 months later with 1 × 10⁶ KPCY cells, and none developed tumors, indicating robust immunological protection.

Using the same KPCY model, we also performed orthotopic tumor implantations in the pancreas (Figure S8). Fifteen days after surgical implantation, mice were randomized into CANDI470 or CANDI^E treatment groups (5 mg of CANDI per mouse; n = 8 per group). At the end of the treatment trial, mice were sacrificed, and the peritoneal cavity was carefully explored. CANDI470-treated mice had substantially smaller tumors compared to empty CANDI^E-treated mice (p = 0.001; Figure 7D). Immunohistochemistry showed much higher levels of CD8⁺ T cell infiltration in CANDI470-treated tumors (Figure S9) and much lower Spp1-positive cell numbers (Figure S10).

Similar therapeutic efficacy was observed in the chemically induced murine Panc02.⁵⁰ On day 42, control tumors had a mean volume of 612.3 ± 102.5 mm³, whereas tumors in CANDI470-treated mice had a mean volume of only 39.7 ± 51.1 mm³ (p < 0.001). Notably, two of five treated animals showed no residual tumor.

To determine whether macrophages in treated tumors had altered levels of Cxcl9 and Spp1 transcriptional expression, we performed quantitative PCR on both murine BMDMs and fresh-frozen murine PDAC tumors treated with CANDI470 (Figure S3). These data showed a time-dependent increase in Cxcl9 and a decrease in Spp1 in CANDI470-treated mice. To further study the cellular effects of CANDI470 in murine PDAC, we performed multiplexed histology on tumor sections of treated and untreated mice. Results from these analyses showed a trend toward increased CD8⁺ T cell infiltration in CANDI470-treated tumors. In contrast, the numbers of CD4 T cells, regulatory T cells, and macrophages were not statistically different between the two groups (Figure S15). In other words, CANDI470 strongly improves the CSR without altering TAM abundance, further indicating that this drug not only promotes the disappearance of tumor-promoting TAM states but also—and simultaneously—the appearance of antitumor TAM states.

To assess potential toxicities, body weight was monitored throughout treatment and remained stable in both CANDI470- and CANDI^E-treated groups (Figure 7D). Histological analysis of the liver and spleen revealed no evidence of organ toxicity (Figures S11 and S12). Blood chemistry and complete blood counts of CANDI470-treated mice showed no statistical differences compared with vehicle-treated control mice (Table S5). These results align with previously reported safety profiles of related CANDI formulations.^47,51

CANDI470 has immune-stimulatory effects in human PDAC tissue

To determine whether CANDI470 elicits similar immune-stimulatory effects in human tissue, we obtained freshly resected human PDAC specimens and generated organotypic tissue slices, which were then cultured ex vivo and exposed to CANDI470. Histological analysis revealed a significant upregulation of myeloid activation markers, including CD86 and MHC-II, in CANDI470-treated tissue sections (Figure S15C). Culture supernatants were subsequently tested for CXCL9 and SPP1 production, showing a dose-dependent increase in CXCL9 production accompanied by a reduction in SPP1 (p < 0.001 at 24 h; Figures S15D and S13), similar trends to those seen in Figure 6. Confocal imaging further showed that CANDI470 localized primarily to CD68⁺ TAMs (Figure S14), corroborating the observations in Figures 5C and 5D.

Discussion

The TME in pancreatic cancer has proven to be a formidable barrier to efficient therapy. The pancreatic TME is composed of numerous populations of largely suppressive immune cells, fibroblasts, a dense extracellular matrix, and often poor vascularity. The neoplastic cells in PDAC themselves often constitute only a minority of the overall cellularity within the tumor, often <10%–20% of total cells,¹⁰ but this can be highly variable.^25,52 Myeloid cells comprise the largest population of immune cells in PDAC. This has prompted investigations into the roles of various immunosuppressive populations.^53,54,55 Targeting various aspects of the innate immune response has been undertaken.^10,18,56,57 This includes immune checkpoint inhibition,^53,54,55,58 agonists for CD40,^59,60,61 and agonists for the CD11b/CD18 receptor,^62,63 among others. Still, while therapeutic responses have been modest, they clearly show enhanced T cell infiltration in PDAC.^60,61

Based on immune cell profiling in different cancers,¹² unique immunosuppressive TAM phenotypes have been identified. Immunosuppressive TAMs that facilitate tumor progression are often characterized by the expression of SPP1 and C1Qc and the lack of IFN-stimulated genes such as CXCL9, CXCL10, and ISG15.^{12,15,64,65,66} It has also been shown that higher IFN-TAM frequencies correlate with increased patient survival and immune checkpoint blockade responses.^12,64,65,67 The mechanisms regulating the differentiation, expansion, and maintenance of IFN-TAMs are slowly being unraveled.⁶⁸ Apart from our work showing pharmacological manipulation,^26,27,47 other studies have demonstrated that extracellular cues from the TME regulate CXCL9/SPP1 TAM polarity (CSR) and functions by modulating the levels of the cytochrome c oxidase (complex IV, CIV) subunit NDUFA4.⁶⁸ TAMs with elevated levels of either CXCL9 or SPP1 occupy nearly mutually exclusive tumor areas, suggesting that the transcriptional programs associated with CXCL9 and SPP1 are determined by local microenvironmental cues.

In the current study, we utilized a 17 nm nanoparticle platform (CANDI) to deliver a cocktail of small-molecule immunomodulatory drugs to TAMs. CANDI by itself is made up of lysine cross-linked bis-succinyl cyclodextrin units, a composition that has been shown to effectively accumulate in TAMs following systemic administration,⁶⁹ similar to large-molecular-weight dextrans often used for macrophage identification.^70,71 Our data confirm the TAM accumulation of CANDI in PDAC, specifically in myeloid cells within the TME. To modulate the TAM phenotype, we loaded three therapeutic payloads (R848, MSA-2p, and PLX3397) into CANDI470, a cocktail that had previously been identified through preclinical screens.^26,27,47 What remained untested, however, was whether this paradigm would be effective in PDAC—specifically, whether it would elicit meaningful immune responses and confer therapeutic benefit.

Here, we show robust local induction of CXCL9 and suppression of SPP1, changes associated with reduced tumor growth and, in many cases, complete tumor regression. CANDI470 treatment was accompanied by marked upregulation of ISGs, including Ifit2, Isg15, and Rsad2, which have been linked to antitumor immunity.⁷² In parallel, we observed downregulation of immunoinhibitory pathways, consistent with a shift toward a more immunostimulatory myeloid state. Notably, CANDI470 reduced expression of Trem2, a receptor associated with immunosuppressive macrophage programs and tumor promotion,^{73,74,75,76,77} while increasing expression of Cd40 and Il12b, two factors with well-established roles in promoting antitumor immune responses.⁷⁸ Our data indicate that CANDI470 preferentially accumulates in TAMs, consistent with prior observations that related CANDI formulations localize to macrophages, potentially through clathrin-associated uptake pathways.⁵¹ The cellular uptake mechanism had previously been studied.⁴⁷ Using inhibitors of different cellular uptake mechanisms, we observed the largest effects with fucoidan-mediated scavenger receptor inhibition, pegging this as the primary uptake mechanism.

CD8⁺ T cell infiltration into tumors occurs when a myeloid environment receptive to T cells is present. Typically, in PDAC, such an environment does not exist, explaining the poor CD8⁺ T cell infiltration and the “immune-cold” status of PDAC tumors. This underlies the lack of response to anti-PD-1 immune checkpoint inhibitors in PDAC. CXCL9 is a well-known factor driving T cell infiltration into tumors.⁷⁹ Our group has previously defined ratios of macrophage CXCL9 and Spp1 production as strong determinants of both T cell-inflamed tumor microenvironments and the response to immunotherapy.¹² As CXCL9 directly recruits T cells into the tumor, our CANDI470 approach is designed to create myeloid environments that are receptive to T cells. Importantly, the durable protection observed in tumor re-challenge experiments indicates that CANDI470 can induce long-lasting antitumor immunity, highlighting its promise as an immunotherapeutic strategy for PDAC.

Our combination therapy includes both the TLR7/8 agonist R848 and the non-nucleotide STING agonist MSA-2 (in a prodrug form to enable loading into CANDI nanoparticles). Prior reports demonstrate that concurrent treatment with R848 and STING agonists can induce synergistic immune stimulation.⁸⁰ R848 and MSA-2 engage non-overlapping immune-stimulatory pathways, and recent investigations demonstrate synergy between TLR pathways and STING signaling through NF-κB pathway activation.^81,82 These studies demonstrate that generalized NF-κB pathway activation leads to heightened STING signaling, which could explain the durable protection observed in tumor re-challenge experiments. These findings indicate that CANDI470 can induce long-lasting antitumor immunity, highlighting its promise as an immunotherapeutic strategy for PDAC.

Our long-term objective is to advance CANDI-based therapeutics into clinical testing for hard-to-treat cancers. While we have demonstrated proof of principle for reprogramming relevant human myeloid cells within the PDAC TME, several steps are required for translation. These include scaling up under GMP/GLP practices, treatment synergy studies, comprehensive toxicity and safety studies, and possibly additional mechanistic investigations. In prior studies, we have shown synergy between certain CANDI formulations and anti-PD1,⁴⁷ radiation therapy,⁵¹ or nab-paclitaxel,⁷ but additional studies with Kras inhibitors, FOLFIRINOX, gemcitabine + nab-paclitaxel, anti-PD1, and radiation would be warranted in PDAC models. Nonetheless, the robustness of the preclinical data presented here strengthens the rationale for advancing CANDI470 and next-generation CANDI formulations toward Investigational New Drug (IND)-enabling studies, with the ultimate goal of testing them in patients with PDAC.

Limitations of the study

A limitation of this study is that in vitro drug treatment experiments were performed using BMDMs, which may differ from bona fide TAMs in phenotype, ontogeny, and therapeutic responsiveness, as they are not conditioned within the PDAC TME. Nevertheless, these experiments were necessary to obtain sufficient numbers of cells for dose-response studies. In addition, while the CANDI470 immunotherapy incorporates the STING agonist MSA-2p, the precise cellular compartment through which STING activation remodels the CSR remains incompletely defined. Future studies using cell-type-specific STING perturbation approaches will be important to more precisely resolve how STING signaling regulates macrophage state transitions within tumors. Nonetheless, the biodistribution of CANDI470 strongly favors macrophages and DCs, and the CSR can be directly shifted in macrophage monoculture systems, collectively suggesting that macrophage-intrinsic STING signaling plays a central role in this process. Broader indicators of IFN-γ responses were likewise not comprehensively characterized, although CXCL9 expression is tightly linked to IFN-γ signaling. Future studies could therefore define the spatial organization of IFN-γ production relative to CXCL9- and SPP1-expressing myeloid populations in vivo. This remains technically challenging for intravital imaging studies because these fluorescent reporters currently exist in separate genetically engineered mouse strains with incompatible fluorophore combinations (IFN-γ-eYFP, CXCL9-RFP, and Spp1-tdTomato). Dynamic spatial mapping of IFN-γ-producing cells, and determining whether CXCL9 induction in vivo is IFN-γ dependent, will be important for defining how macrophage CSR remodeling shapes T cell phenotypes and antitumor immunity in cancer.^83,84,85

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Ralph Weissleder (rweissleder@mgh.harvard.edu).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

Data and code availability

• •The RNA-seq data from the BMDM experiment are deposited at GEO (https://www.ncbi.nlm.nih.gov/geo/) as GEO: GSE334154 and are publicly available as of the date of publication. All microscopy and flow cytometry data reported in this paper will be shared by the lead contact upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

Acknowledgments

We thank Drs. Elias Halabi and Hyung Shik Kim for their assistance with earlier studies on CANDI synthesis, Moonhyun Choi for performing some of the early EM experiments, and Vidhi Sehgal for assisting with the BMDM-Cxcl9 experiment. We thank Drs. Miles Miller, Jonathan Carlson, and Cesar Castro for helpful discussions. This work was supported by the CSB Development Fund and in part by the following NIH grants: R33CA277820, R01CA257623, and R01CA281735. Y.T.M.N., S.A.W., R.J.D.-W., and G.G.S. were supported in part by T32CA079443.

Author contributions

Conceptualization, M.J.P. and R.W.; methodology, G.G.S., J.O., S.A.W., and M.J.P.; formal analysis, Y.T.M.N., M.P., P.W., C.V., J.O., S.N., and R.W.; investigation, Y.T.M.N., G.G.S., J.O., S.A.W., and R.J.D.-W.; resources, R.W.; data curation, M.P., J.O., S.N., S.A.W., C.V., and Y.T.M.N.; writing – original draft, R.W.; writing – review and editing, all co-authors; visualization, M.J.P. and R.W.; supervision, R.W.; project administration, R.W., F.D.S., A.S.L., and C.F.d.C.; funding acquisition, R.W.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-mouse CD45 (clone 30-F11)	R&D Systems	MAB114; RRID:AB_357485
Anti-mouseCD4 (clone D7D2Z)	Cell Signaling Technology	27520; RRID:AB_2798898
Anti-mouse CD8 (clone D4W2Z)	Cell Signaling Technology	60168; RRID:AB_2756376
Anti-mouse FoxP3 (clone FJK-16s)	Thermo Fisher Scientific	14-5773-80; RRID:AB_467575
Anti-mouse F4/80 (clone BM8)	BioLegend	123101; RRID:AB_893504
Anti-mouse SPP1 (clone EPR21138)	Abcam	ab229854; RRID:AB_2732079
Anti-human CD45 (clone D9M8I)	Cell Signaling Technology	47937; RRID:AB_2922773
Anti-human CD68 (clone KP1)	BioLegend	916104; RRID:AB_2616797
Anti-human CD86 (polyclonal)	R&D Systems	AF-141; RRID:AB_354385
Anti-human HLA-DR (clone 1083110)	R&D Systems	MAB11555
Anti-human SPP1 (clone OSP/4589)	Thermo Fisher Scientific	6696-RTM1-P0
Anti-rabbit secondary antibody (Rabbit IgG, AF647)	Abcam	Ab150083; RRID:AB_2714032
Anti-rat secondary antibody (Rat IgM, DyLight755)	Thermo Fisher Scientific	SA5-10015;RRID:AB_2556595
Anti-goat secondary antibody (Goat IgG, AF488)	Thermo Fisher Scientific	A32814TR; RRID:AB_2866497
Anti-CD45, BV421 (clone 30-F11)	BioLegend	103133; RRID:AB_10899570
Anti-CD19, PE-Cy5 (clone 6D5)	BioLegend	115509; RRID:AB_313644
Anti-CD11b, BV650 (clone M1/70)	Biolegend	101259; RRID:AB_2566568
Anti-F4/80, PE-Cy7 (clone BM8)	Biolegend	123114; RRID:AB_893478
Anti-CD11c, PE (clone N418)	Biolegend	117307; RRID:AB_313776
Anti-NKp46, BV605 (clone 29A1.4)	Biolegend	137619; RRID:AB_2562452
Anti-B220, PE (clone RA3-6B2)	BD Pharm	553089; RRID:AB_394619
Anti-CD3, APC-Cy7 (clone 17A2)	Biolegend	100221; RRID:AB_2057374
Anti-TCRbeta, PE (clone H57-597)	Biolegend	109207; RRID:AB_313430
Biological samples
Human PDAC samples	MGH	This study
Chemicals, peptides, and recombinant proteins
Live/Dead Fixable Aqua	Thermo Fisher Scientific	L34966
Resiquimod (R848)	MedChemExpress	HY-13740
Pexidartinib (PLX3397)	MedChemExpress	HY-16749
MSA-2p	MGH, CSB	N/A
L-Lysine	Millipore Sigma	L5501
Succinyl-β-cyclodextrin	Sigma Aldrich	85990
Critical commercial assays
Human CXCL9 ELISA Kit	Abcam	ab100595
Human Osteopontin ELISA Kit	Abcam	ab100618
Deposited data
BMDM RNA-seq dataset	This study	GEO: GSE334154
Experimental models: Cell lines
KPCY: KRAS murine PDAC model	Kerafast	2838c3
KPC-H2B-mTAG: Histone transfected blue fluorescent KPC	CSB, MGB	N/A
Panc02: chemically induced PDAC in C57BL/6	DTC, NIH	6692374
iMACs: Immortalized murine bone marrow-derived macrophages	Charles L. Evavold, MGB	34289345
BMDM: bone-marrow derived macrophages	This study	N/A
Experimental models: Organisms/strains
C57BL/6J	Jackson Laboratory	Stock# 000664 RRID:IMSR_JAX:000664
C57BL/6J-Tg(Cxcl9-RFP,Cxcl10-BFP)1Adl/J	Andrew Luster, MGH	23123063
C57BL/6J-Spp1em1Msasn/J	Jackson Laboratory	Strain #:033731 RRID:IMSR_JAX:033731
Software and algorithms
GraphPad Prism (v.10.1.1)	GraphPad software	RRID: SCR_002798
BioRender	Science Suite Inc.	https://biorendder.com
R (v.4.15)	R Core Team	https://www.r-project.org/
R package edgeR	Bioconductor	https://www.bioconductor.org
R package limma	Bioconductor	https://www.bioconductor.org
R package pheatmap	CRAN	https://cran.r-project.org
R package tidyverse	CRAN	https://cran.r-project.org

Experimental model and study participant details

Mice

Wild-type C57BL/6 (females, 8–10 weeks) were purchased from Jackson Laboratory. Reporter mice including Cxcl9-RFP and SPP1-tdTomato mice (also C57BL/6 background, females, 8–10 weeks) were bred at MGH Research Center. All mice were housed in Institutional Animal Care and Use Committee (IACUC) approved facilities (specific pathogen-free conditions with daily monitoring from vivarium technicians and preventative veterinary care when necessary). Experiments were approved by the MGH IACUC and were performed in accordance with MGH IACUC regulations under protocol 2013N000157.

Tumor cells

The mouse pancreatic cancer model KPCY (2838c3) was acquired from Kerafast, and Panc02 cell lines were from ATCC. To allow tumor demarcation with the Spp1-tdTomato/MerTK-GFP reporter mouse model during intravital imaging, non-fluorescent 2838c3 cells were first sorted using an Aria 3/FACSAriaII. Then, the cells were transduced with a lentiviral vector encoding H2B-BFP (pLVX-BFP-Puromycin; Clontech) in the presence of Polybrene (10 μg/mL, Santa Cruz Biotechnology). Transduced cells were subsequently selected using puromycin (3 μg/mL).

All cell lines were maintained under standard culture conditions. Specifically, Dulbecco’s Modified Eagle Medium (DMEM; Corning) was supplemented with 10% fetal bovine serum (FBS; Corning) and 1% penicillin-streptomycin (Corning) at 37°C in a humidified atmosphere with 5% CO₂. Cells were passaged using 0.05% trypsin/0.53 mM EDTA (Corning) once they reached full confluency. All experiments were conducted using cells at approximately 90% confluency.

Bone marrow-derived macrophages (BMDM)

Bone marrow-derived macrophages were isolated from wild-type, Cxcl9-RFP/Cxcl10-BFP reporter mice, or Spp1-tdTomato/MerTK-GFP reporter mice (see below). Briefly, femurs were excised and flushed with sterile PBS using a 28-gauge needle and syringe to harvest whole bone marrow. Red blood cells were removed using RBC Lysis Buffer (BioLegend) following the manufacturer’s protocol. The resulting cell suspension was counted using a Neubauer hemocytometer and plated at a density of 0.8 × 10⁵ cells per well into glass-bottom 96-well plates (Ibidi, for imaging). To differentiate cells, they were cultured for 7 days in media supplemented with 50 ng/mL recombinant murine M-CSF (BioLegend), with media changes every 3–4 days.

Immortalized macrophage cell line

Immortalized murine bone marrow-derived macrophages (iMACs), kindly provided by Charles L. Evavold (Ragone Institute, Harvard University), and used for gene transfection and cytotoxicity assays. iMACs were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Corning) supplemented with 10% fetal bovine serum (FBS, Corning) and 1% penicillin-streptomycin (Corning) at 37°C and 5% CO₂.

Human samples

Clinical PDAC samples was obtained from patients following the established Institutional Review Board protocol (IRB 2003P001289) at Mass General Hospital. Briefly, informed consents were obtained from patients with formal PDAC diagnosis (male and female, >20 years old, physiologically fit for surgical resection, with or without prior standard chemo-radio therapy treatment, and not under any immune suppressant medications). Freshly resected tissue was collected by surgeons at the Division of Gastrointestinal and Oncologic Surgery and later assessed by pathologists. Upon reconfirming the tumor status, sample was immediately placed in cold 1xPBS and transferred to the lab, where further processing was performed (see human organotypic tissue slices section).

Method details

Synthesis and characterization

Particle synthesis

The β-cyclodextrin (β-CD) core was first functionalized with succinyl groups following established protocols,⁴⁷ yielding succinylated β-cyclodextrin (s-β-CD) with an average degree of substitution (DS) of 2.5 succinyl groups per molecule. For core particle assembly, s-β-CD (DS 2.5, 10 mg, 6.99 μmol) was activated by reaction with EDC (1.5 g, 10 equivalents to carboxylate) and NHS (550 mg, 5 equivalents) in 6 mL of MES buffer (50 mM, pH 6.5). Subsequently, a solution of l-lysine (35 mg, 0.25 equivalents to carboxylate) in MES buffer (1.5 mL) was added in a dropwise manner over 30 min using a syringe pump, and the reaction was allowed to stir for 18 h at 25°C. The particles were precipitated by slow addition over 15 min into absolute ethanol (70 mL) cooled to −10°C with a brine/ice bath, yielding a white precipitate that was decanted and dissolved in water (14 mL). The aqueous particle solution was purified with 10 kDa MWCO centrifugal filters (Amicon; 10,000 g for 8 min) and lyophilized for 48 h. The dry particles (∼320 mg) were characterized by DLS (2 mg/mL) and zeta potential (2 mg/mL) and stored as solids at −20°C until further use.

MSA-2p precursor synthesis

The pseudo-lactone pro-drug version of MSA-2, termed MSA-2p, was previously developed for incorporation into cyclodextrins.²⁸ This synthesis, however, was not easily scalable and expensive. In an effort to more economically furnish large quantities of MSA-2p (Figure S2), the synthesis was begun with the 5,6-dimethoxybenzo{b}thiophene-2-carboxylic acid, which—after a brief screen of conditions—was decarboxylated with catalytic silver acetate and potassium carbonate in quantitative yields and easily purified. Specifically, into an oven-dried round-bottom flask equipped with a magnetic stirrer, silver (I) acetate (105.1 mg, 629.6 μmol, 0.1 equiv.), potassium carbonate (130.5 mg, 944.4 μmol, 0.15 equiv.), and 5,6-dimethoxybenzo{b}thiophene-2-carboxylic acid (1.50 g, 6.30 mmol, 1.0 equivalent) were added, the flask was evacuated and backfilled with argon ×3. To the solids under argon, 12.6 mL (0.5 M) of N-methyl-2-pyrrolidone (NMP) was added, and the reaction was heated to 120°C, resulting in a rapid color change from clear yellow to a purple and rust-red solution, and thereafter gradually became an opaque matte gray color. The reaction was monitored by LCMS for completion, and after 18 h the reaction was cooled. The reaction was diluted with DCM and 5N HCl, and extracted with DCM ×3. The DCM layers were combined, washed with sat. aq. K₂CO₃ solution. To remove residual NMP, the DCM was reduced to dryness by rotary evaporation, resuspended in ethyl acetate (200 mL), and washed with copious (900 mL) of 5% LiCl solution ×1 and saturated brine solution ×5. The organic layer was dried over Na₂SO₄ and evaporated to dryness (40°C), resulting in a solid in quantitative yields (1.22 g, 6.30 mmol); where necessary, the solids were purified via a silica plug.

Small molecule loading of nanoparticles

The drug loading for the triple combination chosen here is ∼0.1 mg total drug per mg nanoparticle, which is classified as a high drug loading (>10 wt %) nanoparticle.⁸⁶ For other drugs the payload loading has been determined to be ∼0.2 mg/mg.⁴⁷ The composition and ratios of the payloads had been defined by prior screening experiments where different combinations were tested in BMDM from reporter mice^26,27,28 to induce desirable signals, such as Cxcl9, and to reduce undesirable signals such as Spp1. To prepare the drug-loaded CANDI470, 5 mg of empty CANDI was dissolved in 90 μL of 1x PBS solution. Then, 10 μL of DMSO containing payloads (0.20 mg of MSA-2p, 0.112 mg of Pexidartinib, and 0.176 mg of R848) was added to the empty CANDI solution. The mixture was then vortexed vigorously for 3–5 min until the drugs were completely dissolved. All solutions were filtered through a 0.22 μm sterile filter (VWR) prior to in vitro and in vivo studies.

Nanoparticle stability evaluation

To assess the stability of newly prepared or freshly thawed CANDI470, physicochemical features, including hydrodynamic radius and polydispersity index (PDI), were measured using dynamic light scattering (DLS; Malvern Zetasizer Nano ZS) over 3 days. To assess the feasibility of preparing CANDI470 in advance of repeated in vivo animal treatments, CANDI470 was prepared as stated above at a larger scale (>50 mg), then placed into multiple aliquots. The solution is then immediately submerged in liquid nitrogen. Frozen aliquots were thawed and immediately evaluated for size and PDI over a period of 2 weeks, 1 month, 3 months, 4 months and 6 months to assess their stability. Consistent particle size, PDI <0.3, and particle size were considered stable and suitable for injectable applications.

Particle size and charge

The hydrodynamic diameter and zeta potential of nanoparticles were measured using dynamic light scattering (DLS) and electrophoretic light scattering (Zetasizer APS, Malvern Instruments). For these assessments, samples were prepared at a concentration of 1–3 mg/mL in MilliQ water or PBS. All measurements were carried out at 25°C using disposable cuvettes (Malvern Instruments).

Electron microscopy

Freshly prepared CANDI nanoparticles (5 mg/mL in MilliQ water) were diluted to 1 mg/mL with ultrapure water. A drop of the diluted suspension was deposited onto a transmission electron microscopy (TEM) grid and allowed to adsorb for 1 min. The grid was then stained with 2% (w/v) uranyl acetate for 15 min, followed by three washes with ultrapure water. Imaging was performed using a JEOL 2100 transmission electron microscope.

Drug release kinetics

The release kinetics of CANDI470 payloads were evaluated using a closed dialysis system equipped with a 3.5–5 kDa molecular weight cutoff membrane (Pur-A-Lyzer Midi Dialysis Kit, Sigma-Aldrich). Solutions of CANDI470 (containing 50 mg CANDI complexed with 1.76 mg of R848, 1.12 mg of Pexidartinib and 2 mg of MSA-2p) were prepared to final volumes of 1 mL of 1× PBS and 10% DMSO. The solution is placed inside the dialysis device, which was immersed in 5 mL of 1× PBS at 37°C under continuous agitation (600 rpm). At designated time points, 100 μL aliquots were collected from the external buffer to assess payload release. For every 100 μL aliquot removed, 100 μL of 1x PBS would be added to the inner chamber. The percentage of eluted molecules was quantified by analysis of the liquid chromatograph-mass spectrophotometer (LC-MS) at different time points (t = 0, 3, 10, 20, and 30 min; then 1, 1.5, 3, 4, 5, 6, 7, 8, and 9 h). Each payload was identified by its distinctive retention time (R848 = 0.67 min, MSA-2p = 1.18 min, and Pexidartinib = 0.89 min) and mass-to-charge ratio (ES−: R848 = 316, MSA-2p = 309, and Pexidartinib = 420). The cumulative drug release was determined as the ratio of the integrated area under the curve for each eluted peak to the total area under the curve of chromatographs obtained from the non-dialyzed solutions. All experiments were conducted in triplicate (n = 3).

Cell experiments

Live-cell imaging of Cxcl9-BFP and Spp1-tdTomato expression in bone marrow-derived macrophages

Bone marrow-derived macrophages from reporter mice were seeded in 96-well plates (50,000 cell per well) or 8-well plates (85,000 cells per well). Cells were treated with CANDI470 (1 mg/mL) twice. Once on day 4 of differentiation for 3 h. Then, an overnight treatment was given on day 6. Live-cell fluorescence imaging was performed on day 7 using an IX81 inverted fluorescence microscope (Olympus, Tokyo, Japan) equipped with a motorized stage (Renishaw, Wotton-under-Edge, UK) and an ORCA-Fusion Digital CMOS camera (Hamamatsu Photonics, Hamamatsu, Japan). Images were acquired using CellSens Dimension 3.1.1 software (Olympus) with either a UPlanSApo ×10 objective (NA 0.75) or a UPlanSApo ×40 air objective (NA 0.95). Brightfield images and fluorescence channels corresponding to BFP (excitation/emission: 345/455 nm) and tdTomato (554/581 nm) were collected using appropriate optical filters. Fiji and Cell Profiler were used to analyze the image data.

For cytotoxicity analysis, iMACs were seeded into 96-well plates at a density of 1.5 × 10⁴ cells per well and incubated for 24 h at 37°C in a humidified 5% CO₂ atmosphere. CANDI470 stock solution (10 mg/mL) was prepared by first dissolving 10 mg of CANDI in 100 μL of complete DMEM. Payloads (0.40 mg MSA-2p, 0.35 mg R848, and 0.22 mg PLX3397) were added and vortexed rigorously before additional culture media DMEM was placed into the final volume of 1 mL. This solution is serially diluted in complete culture medium to final concentrations ranging from 10 mg/mL, 5 mg/mL, 2.5 mg/mL, 1.25 mg/mL, 1.0 mg/mL, 0.5 mg/mL, 0.25 mg/mL, 0.125 mg/mL, 0.05 mg/mL, 0.025 mg/mL, and 0.005 mg/mL. Negative control was given no CANDI. Final DMSO concentration in all treatment solutions was 0.5%. Cells were treated for 2.5 h, then replaced with fresh media. After an additional 48 h of culturing, cell viability was assessed using Alamar Blue (Thermofisher) and TECAN microplate reader.

RNA-seq analysis

BMDMs were cultured as described. On day 7 after differentiation, CANDI470 or empty CANDI was added to the final concentration of 1 mg/mL. RNA was extracted using Qiagen RNeasy Kit per instructions 12 h after the treatment. RNA samples were sent to Novogene for RNA integrity testing, mRNA library preparation, and sequencing (NovaSeq X Plus Series, PE150). From the count data, differentially expressed genes between the control and the empty or CANDI 470 were identified using the edgeR and limma packages under R version 4.4.3.Interferon-gamma–responsive genes were extracted from the HALLMARK_INTERFERON_GAMMA_RESPONSE gene set in the MSigDB database, and the top 20 genes showing the largest expression changes were visualized in the heatmap.

Quantitative PCR

To validate the observed change in RNA-seq and to evaluate the transcriptional modulation of Cxcl9 and Spp1, BMDM were cultured and differentiated (3 x 10⁶ cells per 10 cm TC-plate), as described. Cells were treated with CANDI470 (1 mg/mL) twice. Once on day 4 for 3 h, then replaced with fresh media. An overnight treatment was applied on day 6. RNA was extracted on day 7 using RNeasy kit from Qiagen. qPCR assay was performed using mouse-specific primers designed by Origene (SPP1_cat# MP215080, CXCL9_cat# MP202421, and GAPDH_cat# MP205604).

In vivo experiments

Tumor Cell implantation

For tumor engraftment studies, KPCY 2838c3 cells (0.3 × 10⁶ cells per mouse) or Panc02 cells (1 × 10⁶ cells per mouse) were subcutaneously injected into the flank region of C57BL/6J mice. Tumor development was monitored regularly, and therapeutic interventions commenced after one week, once tumors reached a minimum size of approximately 30 mm³. For tumor re-challenge experiments, 1 × 10⁶ KPCY cells were injected into the same flank previously used for primary tumor inoculation.

Biodistribution experiments

For biodistribution analysis, 2838c3-bearing C57BL/6J mice (n = 3) were administered AF647-labeled CANDI470 via intravenous injection through the tail vein. After overnight, in vivo fluorescence imaging was performed using the IVIS Spectrum Imaging System (PerkinElmer) under anesthesia with 2% isoflurane delivered in 2 L/min O₂ (VetEquip). Following imaging, mice were euthanized, and major organs (including liver, spleen, lungs, kidneys, heart, and tumors) were harvested, weighed, and subjected to ex vivo fluorescence imaging. Quantitative analysis of fluorescence signals was performed by defining regions of interest (ROI) for each organ, and the fluorescence intensity was normalized to tissue weight (photons/s/cm²/sr/mg) using Living Image Software (PerkinElmer).

Intravital imaging

Dorsal skinfold window chambers were surgically implanted with minor modifications based on well established protocols. Intravital confocal imaging was performed using a customized Olympus FV1000 confocal microscope (Olympus America). Imaging utilized 2× (XLFluor, NA 0.14), 4× (UPlanSApo, NA 0.16), and 20× (XLUMPlanFL N, NA 1.0) water immersion objectives (Olympus America). KPCY tumor cells or KPC-H2B-BFP, AF647-labeled CANDI470 (5 mg), and vascular probes were sequentially excited using 405 nm, 473 nm, 559 nm, and 633 nm diode lasers, respectively, coupled with a DM-405/488/559/635 nm multiband dichroic beam splitter. Emitted fluorescence was spectrally separated using beam splitters (SDM-473, SDM-560, SDM-640) and corresponding emission filters (BA430-455, BA490-540, BA575-620, and BA655-755; Olympus America). Laser power settings were carefully adjusted to minimize photobleaching, phototoxicity, and tissue damage. Acquired images were processed and analyzed using Fiji software (ImageJ2, Version 2.3/1.53f).

CANDI470 distribution in murine tumor

KPCY 2838c3 tumor-bearing C57BL/6J mice were administered a single intravenous dose of AF647-labeled CANDI470. Tumors were harvested 18 h post-injection. Tissues were enzymatically dissociated in RPMI 1640 medium supplemented with 0.2 mg/mL collagenase D (Roche) at 37°C, followed by mechanical dissociation and filtration through 70 μm cell strainers (Corning) to obtain single-cell suspensions. Viability was assessed using AquaAmine Live/Dead Fixable Viability Dye (Thermo Fisher Scientific) according to the manufacturer’s protocol. After washing, cells were incubated with an Fc receptor-blocking antibody (anti-mouse CD16/32, BioLegend) to minimize nonspecific binding, followed by surface staining with fluorochrome-conjugated antibodies, as listed in Table S1. Antibodies were diluted in FACS buffer (1× PBS supplemented with 2 mM EDTA and 2% FBS). The uptake of CANDI470-AF647 by various immune cell subsets was quantified. Samples were acquired in triplicate using the Attune NxT Flow Cytometer (Thermo Fisher Scientific) and analyzed using FlowJo v10 software (TreeStar).

Efficacy studies

A series of therapeutic efficacy studies were performed. The dose of drug product was kept constant across the studies (per mouse, 5 mg of CANDI was loaded with active drug R848 (0.176 mg), pexidartinib (0.112 mg) and MSA2p (0.20 mg)) and was based on prior efficacy studies^26,27,28 and projected human doses.^87,88 Mice were implanted with murine-derived syngeneic tumor cells, KPCY 2838c3 or Panc02, either ectopically or orthotopically into the pancreas. When the tumor became detectable (∼50–80 mm³ via caliper measurements in the ectopic model, or palpable in the orthotopic model), mice were randomized into control (CANDI^E) or treatment (CANDI470) groups. Animals in both the control and treatment groups received four intravenous injections, spaced 4 days apart. In the ectopic model, tumor burden was continuously monitored, and a tumor volume of 1000 mm³ was used as the endpoint for survival analysis. Tumor regressions were calculated as the difference between the initial tumor size at the start of treatment and the tumor size at harvest. In the orthotopic model, surgical exploration was performed to harvest tumors and weigh them in the two groups.

Histology

Frozen tissue sections were cut to a thickness of 5 μm and then processed for immunofluorescence. Frozen tissue sections were thawed, rehydrated, and blocked with Intercept Blocking buffer (LI-COR) for 30 min before antibody staining. Tissue sections were incubated with antibodies (Table S3) for 1–2 h and washed three times with PBS for 10 min each. An Olympus BX-63 microscope was used for image acquisition (Metamorph software version 7.10.4). CellProfiler version 4.1.8 was used for analyzing tissue section images. A custom Python script was used to generate synthetic image maps.

Toxicity experiments

Wild-type C57BL/6J mice (n = 3 per group) were injected with different drugs via the tail vein. Twenty-four hours later, mice were sacrificed. Blood was drawn for comprehensive testing (serum analytes, blood counts, metabolic parameters) (Table S5). Mice were then perfused with PBS through the left ventricle before the collection of the liver for histology. Livers were fixed in 4% paraformaldehyde overnight before washing in ethanol (70%), embedding in paraffin, and processing for hematoxylin-eosin staining (Figure S11).

Human studies

Bioinformatics analysis

Raw bulk RNA-seq data from TCGA were downloaded and log₂-transformed. Gene-level matrices were generated by converting Ensembl IDs to gene symbols while collapsing rows sharing the same gene symbol by summation. The resulting expression matrix was mean-centered. The CXCL9–SPP1 score (“CSR”) was computed directly in log₂ space as the difference log₂(CXCL9) – log₂(SPP1) (with small offsets added to avoid zero denominators), and transformed to linear scale via exp(CSR) for visualization. Samples were merged with TCGA clinical metadata, and a total of 11 tumor types were selected. Head and neck cancers were grouped into a single category based on tumors originating from the larynx, tonsils, oral cavity, base of tongue, hypopharynx, lip, palate, and related subsites.

Testing in human organotypic tissue slices

To determine the effects of CANDI470 on PDAC clinical sections (n = 4), freshly resected specimens were transferred from the operating room to the laboratory in a 50 mL conical tube containing ice-cold medium. Tissue was sliced on a vibratome to a thickness of 300 μm. PDAC slices were cultured on a PTFE insert (pore size 0.4 μm) in 24-well plates with culture media (Dulbecco’s Modified Eagle Medium supplemented with 10% Fetal Bovine Serum and 1x Pen-Strep solution). Tumor slices were cultured for 1 day following CANDI exposure. Supernatant was collected for cytokine analysis. Tissue slices were then fixed in 4% paraformaldehyde for immunohistofluorescence.

To assess CANDI470 uptake by TAM, whole tumor slices were incubated with CANDI470-AF647 or unlabelled empty CANDI for 24 h. The samples were then stained for CD86 while free-floating (1 day on a rocking platform at 4°C). Confocal imaging was performed using a customized Olympus FV1000 confocal microscope (Olympus America). A 2x (XLFluor, NA 0.14), a 4x (UPlanSApo, NA 0.16), and an XLUMPlanFL N 20x (NA 1.0) water immersion objective were used for imaging (Olympus America).

Quantification and statistical analysis

All statistical analyses were performed using GraphPad Prism 11. For normally-distributed two-group analysis, we used a 2-tailed Student’s t test and one-way ANOVA followed by Bonferroni’s multiple comparison test. When variables were not normally distributed, we performed non-parametric Mann-Whitney or Kruskal-Wallis tests. For multiple comparisons, we used a one-way ANOVA with Dunnett’s post hoc test. The significance of overall survival was assessed using the Kaplan-Meier method and the log rank test. For the in vitro assessment of CANDI470 (Figure 3), we performed nonlinear regression analysis. For in-vivo experiments, sample size was determined using power analysis, whereas the initial pilot experiment (n = 4 per group) informed the effect size and statistical power. Results are represented as mean ± Standard Error of the Mean (SEM). p < 0.05 was considered significant. ∗ indicates p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.0001. n.s. denotes not significant.

Published: July 23, 2026